Amount Awarded: $25,000
Forecasting with Signatures: From Theory to Applications (FORESIG) introduces a mathematically grounded framework for modeling and predicting complex, non-stationary time series. By leveraging iterated-integral signatures and the emerging signature kernel, our interdisciplinary team—Professors Harang (BI Norwegian Business School), Ebrahimi-Fard (NTNU), and Guo (UC Berkeley)—will turn nonlinear dynamics into linear features that are both interpretable and data-efficient.
Over the two-year period (August 2025–July 2027), we will:
- Design and implement regression and kernel methods based on signature transforms to capture high-frequency volatility and multi-dimensional interactions;
- Benchmark these methods against classical (ARIMA, ETS) and modern (LSTM, GRU) approaches using real-world retail datasets from Amazon and Rema 1000;
- Establish theoretical guarantees—consistency proofs, convergence rates, and interpretability analyses—for our signature models and kernels;
- Disseminate our results through an open-source Python library, two international workshops, and comprehensive training materials.
With two industry partners on board if the project is granted (Amazon and REMA 1000), as well as additional budget support totaling a $40 K budget, we will cover travel, lodging, venues, workshops and meetings, as well as a 15 % contingency. FORESIG builds on prior pilot successes—such as multimillion-dollar savings at Amazon—and deep algebraic, analytic, probabilistic and computational expertise. The project promises broadly applicable forecasting solutions that are accurate, transparent, and deployable, catalyzing adoption of signature-based methods across operations research, finance, energy, and beyond. The advancements in the FORESIG project will lay the basis for future grant applications to larger collaborative calls from the EU Horizon, The Norwegian Research Council or the National Science Foundation. A distinct goal of the Peder Sather grant is to end up with a solid sketch for a large grant proposal that will be submitted by end of 2027.